Can AI Help Improve Atherosclerotic Plaque Quantification?

by | Aug 23, 2021 | All News, Blog

Imaging of atherosclerotic plaque using coronary computed tomography angiography (CTA) has demonstrated value in estimating patient risk and guiding care decisions. During the 2021 AI and Machine Learning in Cardiovascular CT Series hosted by the Society for Cardiovascular Computed Tomography (SCCT), Dr. Todd Villines noted that the industry is lacking in standardization of plaque quantification and how findings are described. This is an opportunity where AI and other advanced informatics could be applied to coronary CTA images to unlock valuable information on atherosclerotic plaque characteristics that could be used to personalize care decisions.

Atherosclerotic Cardiovascular Disease

Atherosclerosis, also known as atherosclerotic cardiovascular disease, occurs when plaque builds up on the coronary artery walls. Atherosclerotic plaque can be made up of fatty substances, cholesterol, cellular waste, calcium, and fibrin. As plaque accumulates on the walls of the coronary arteries, less blood can flow through the arteries due to the narrowing of the vessels. This can lead to serious conditions, including heart attack, stroke, and even death.

Plaque Imaging by Coronary Computed Tomography Angiography (CTA)

Evidence suggests that imaging atherosclerotic plaque is essential for estimating patient risk and guiding preventive care decisions. Coronary CTA, which is recommended as a first-line diagnostic test when evaluating patients with stable chest pain, can provide valuable clinical information on coronary atherosclerosis and high-risk plaque features in addition to coronary stenosis [1].

In the publication “High-Risk Coronary Plaque on Computed Tomography Angiography: Time to Recognize a New Imaging Risk Factor,” Dr. Maros Ferencik of OHSU and Dr. Udo Hoffman of Massachusetts general Hospital explain that coronary CTA images allow clinicians to accurately assess the presence, morphology, and composition of coronary atherosclerosis, which includes identifying high-risk plaque features such as positive remodeling, low CT attenuation plaque, napkin-ring sign, and spotty calcium [2].

When imaging atherosclerotic plaque on coronary CTA, it is essential to identify non-obstructive plaque in the coronary arteries. Research presented at the American Heart Association’s Quality of Care and Outcomes Research 2014 Scientific Session demonstrated that non-obstructive coronary artery disease—which is characterized by atherosclerotic plaque that is not expected to obstruct blood flow—was associated with a 28 to 44 percent increased risk of major cardiac events.

2021 Guidelines on Coronary CTA Imaging of Atherosclerotic Plaque

To provide standardization on how to identify and quantify plaque features, the SCCT and the North American Society of Cardiovascular Imaging (NASCI) released the “Expert Consensus Document on Coronary CT Imaging of Atherosclerotic Plaque.” The consensus statement provides information on how clinicians can appropriately identify and quantify non-obstructive atherosclerosis and high-risk plaque features on coronary CTA images while detailing how this information can be used in evidence-based clinical decision-making.

In the consensus statement, Leslee J. Shaw et al. provide recommendations for the visualization and measurement of coronary atherosclerotic plaque [1]. This includes:

  • Documentation of the Presence of Atherosclerotic Plaque: Interpreting physicians should include integrated evidence of coronary atherosclerotic plaque into the laboratory standard report
  • Reporting High-Risk Plaque (HRP) Features: The physician should report the presence of HRP when present in patients with evidence of coronary atherosclerotic plaque to improve risk stratification and guide clinical management decisions
  • Utilizing CAD-RADS System for Reporting of Coronary CTA: It is recommended to use the CAD-RADS modifier “V” if a coronary plaque demonstrates 2 or more HRP features in the coronary CTA final interpretation, or to describe what HRP characteristics are present
  • Semi-quantitative Assessment of the Number of Coronary Segments with Plaque using the Segment Involvement Score (SIS): The CT imager should consider including the SIS in the coronary CTA final interpretation
  • Quantification of Coronary Artery Calcium (CAC) Score: It is recommended to report the total CAC score and the associated risk category
  • Writing the Reporting Conclusion: The conclusion of the report should include a statement regarding the overall amount or extent of atherosclerotic plaque

The consensus document above provides a comprehensive review of the available evidence on imaging of coronary atherosclerosis while equipping clinicians with current information on how to communicate this information in clinical practice to improve care for at-risk patients.

Deep Learning-based Atherosclerotic Plaque Analysis

Although many agree that evaluating coronary atherosclerosis on CTA images could inform patient risk and treatment decisions, it is rarely done because the manual process is time-consuming for clinicians. Deep learning algorithms can be applied to coronary CTA images to automatically classify atherosclerotic plaque and deliver analysis results in a standardized way to improve reporting quality and consistency.

Figure 1: Deep learning algorithms provide an accurate and comprehensive evaluation of the coronary arteries using coronary CTA images

Researchers have been investigating how AI techniques can be applied to coronary CTA images to characterize and quantify atherosclerotic plaque for several years. In a study published in IEEE, research led by Dr. Majd Zreik of the Image Sciences Institute at the University Medical Center Utrecht explored how a recurrent convolutional neural network (CNN) could be used to automatically detect and classify coronary artery plaque and stenosis on coronary CTA images. The researchers retrospectively collected 163 coronary CTA exams for use in the study. According to the study results, the algorithm achieved an accuracy of 0.77 for detecting and characterizing coronary plaque, and an accuracy of 0.8 for detection of stenosis and determination of its anatomical significance [3].

More recently, a team of researchers led by Dr. Damini Dey of Cedars-Sinai Medical Center in Los Angeles developed and evaluated the performance of an AI algorithm used to quantify plaque on coronary CTA images. Dr. Andrew Lin of Cedars-Sinai Medical Center presented these findings at the 2021 SCCT Annual Meeting in July. According to Dr. Lin, the deep learning-based method was externally validated with data from 11 sites, which included 921 patients with 5,045 lesions. The study assessed the diagnostic accuracy of the deep learning-based method in automatically measuring plaque volume and stenosis severity as compared to expert readers and intravascular ultrasound exams.

According to the study results, the new technique was consistent with the performance of expert readers, yielding an intraclass correlation coefficient of 0.964 for total plaque volume; 0.949 for non-calcified plaque, 0.945 for calcified plaque, and 0.777 for low-density non-calcified plaque. Dr. Lin also shared that the deep learning-based algorithm had a fast processing time, generating results in about 5.7 seconds. This demonstrates the added value that AI-based techniques could add in automating plaque analysis in clinical practice to help clinicians make better care decisions for their patients.

Overcoming Challenges for Deep Learning-based Plaque Analysis

While research demonstrates that AI can help facilitate the automatic evaluation and characterization of plaque characteristics, several challenges remain. Automatically characterizing non-calcified plaque on coronary CTA using AI has historically been a challenge for developers. According to a review on machine learning-based analysis of coronary artery disease in cardiac CT published in Frontiers in Cardiovascular Medicine, characterizing the individual components in non-calcified plaque is challenging because of the low-contrast boundaries between the components [4]. As a result, much of the research on AI-based methods for atherosclerotic plaque analysis have focused on calcified plaque.

Bringing Automatic Quantification of Atherosclerotic Plaque to Clinical Practice

Dr. Alastair J. Moss and Dr. Michelle C. Williams published a review in the Journal of Cardiovascular Computed Tomography titled, “Can We Measure Vulnerable Plaques on Coronary CT Angiography with Both Precision And Accuracy?” In the publication, Dr. Williams and Dr. Moss highlight research led by Dr. Márton Kolossváry of Semmelweis University Heart Center in Hungary investigating how radiomic feature extraction from coronary CTA could be a potential solution. According to Dr. Kolossváry’s research, radiomics can provide parameters ranging from relatively simple intensity-based metrics to more complex texture-based metrics that appear more robust to the changes in contours provided by automated or manual editing [5,6]. However, it is also acknowledged that the analysis would have to be performed by experts to be accurate. Additional research is necessary to understand how radiomic feature extraction could help improve the measurement of plaques on CTA.

Genomics and proteomics have demonstrated potential in predicting plaque vulnerability in the coronary arteries. A team of Icelandic researchers led by Dr. Vilmundur Gudnason of the Icelandic Heart Association, Tamara Harris of the National Institute on Aging, and Dr. Lenore Launer of the National Institute on Aging are actively conducting the AGES Reykjavik Study to identify new genetic risk factors for diseases and conditions including atherosclerosis. The study will phenotype the surviving 12,000 members of the Reykjavik Study cohort–in which patients were originally included in January 2006–for quantitative traits related to diseases and conditions of old age, and the researchers will collect genetic and other biologic specimens. In July 2021, Alexander Gudjonsson et al. published findings in the research article “A Genome-wide Association Study of Serum Proteins Reveals Shared Loci with Common Diseases.” According to the article, integrating protein measurements with deep phenotyping of the Reykjavik Study cohort has enabled substantial enrichment of phenotype associations for serum proteins regulated by established GWAS loci, while offering additional insights regarding the interplay between genetics, serum protein levels, and complex disease [7].

This is an area where deep learning-based coronary CTA analysis software may be combined with genomics to aid in measuring and predicting coronary atherosclerotic plaque. Interdisciplinary team so researchers can help accelerate progress in understanding how deep learning-based methodologies can help clinicians to unlock valuable information on coronary CTA exams.

References
[1] Shaw LJ, Blankstein R, Bax JJ, Ferencik M, Sommer Bittencourt M, Min JK, Berman DS, Leipsic J, Villines TC, Dey D, Al’aref S, Williams MC, Lin F, Baskaran L, Litt H, Litmanovich D, Cury R, Gianni U, van den Hoogen, I, Van Roxenhael AR, Budoff M, Chang HJ, Hecht HE, Feuchtner G, Ahmadi A, Ghoshajra BB, Newby D, Chandrashekhar YS, Narula J. Society of Cardiovascular Computed Tomography / North American Society of Cardiovascular Imaging – Expert Consensus Document on Coronary CT Imaging of Athersclerotic Plaque. Journal of Cardiovascular Computed Tomography. 2021 Mar;15(2):93-109. doi: https://doi.org/10.1016/j.jcct.2020.11.002

[2] Ferencik M & Hoffmann U, “High-Risk Coronary Plaque on Computed Tomography Angiography: Time to Recognize a New Imaging Risk Factor,” in Circulation: Cardiovascular Imaging, 2018;11:e007288

[3] M. Zreik, R. W. van Hamersvelt, J. M. Wolterink, T. Leiner, M. A. Viergever and I. Išgum, “A Recurrent CNN for Automatic Detection and Classification of Coronary Artery Plaque and Stenosis in Coronary CT Angiography,” in IEEE Transactions on Medical Imaging, vol. 38, no. 7, pp. 1588-1598, July 2019, doi: 10.1109/TMI.2018.2883807.

[4] Hampe N, Wolterink JM, van Velzen SGM, Leiner T and Išgum I (2019) Machine Learning for Assessment of Coronary Artery Disease in Cardiac CT: A Survey. Front. Cardiovasc. Med. 6:172. doi: 10.3389/fcvm.2019.00172

[5] Moss AJ & Williams MC. Can We Measure Vulnerable Plaques on Coronary CT Angiography with Both. Precision and Accuracy? Journal of Cardiovascular Computed Tomography. 15(2), pp. 146-147. DOI:https://doi.org/10.1016/j.jcct.2020.08.007

[6] Kolossvary M, Javorszky N, Karady J, Vescey-Nagy M, David TZ, et al. Effect of vessel wall segmentation on volumetric and radiomic paramters of coronary plaques with adverse characteristics. Journal of Cardiovascular Computed Tomography. 2020.

[7] Gudjonsson A, Gudmundsdottir V, Axelsson GT, Gudmundsson EF, Jonsson BG, Launer LJ, Lamb JR, Jennings LL, Aspelund T, Emilsson V, Gudnason V. A genome-wide association study of serum proteins reveals shared loci with common diseases. BioRxiv 2021.07.02.450858; doi: https://doi.org/10.1101/2021.07.02.450858